Results

Corpus overview

The pipeline currently covers papers and iGEM projects. Patents and parts will be added in a later pipeline run.

Artifact type	Count	Year range	Cities geocoded
Papers	9,449	2000–2026	8,720 (92%)
iGEM projects	4,615	2009–2025	4,615 (100%)
Patents	—	—	—
iGEM parts	—	—	—
Total	14,064	2000–2026	13,335 (95%)

Papers were retrieved from OpenAlex using core synthetic biology keywords (“synthetic biology”, “synthetic genomics”, “BioBrick”) and subfield keywords (“repressilator”, “minimal genome”, “xenobiology”, etc.). iGEM projects were geocoded using a three-tier pipeline: OpenAlex institution lookup → Claude Haiku city extraction → Nominatim coordinates. Paper coordinates were obtained by fetching full institution objects from OpenAlex (which include city and lat/lon) using institution IDs stored during ingestion.

Semantic space

All artifacts were embedded using SPECTER (Lo et al., 2020), a transformer model pre-trained on scientific text to produce 768-dimensional vectors from title and abstract. Embeddings were reduced to 2D with UMAP for visualization.

The shared projection shows papers and iGEM projects occupying overlapping but distinguishable regions of semantic space, consistent with the hypothesis that student projects and academic research share the same knowledge domain.

See the Explorer for the interactive version.

Clustering

HDBSCAN was applied to the full 768-dimensional embedding space.

	Count
Total artifacts	14,064
Clusters found	256
Artifacts assigned to a cluster	9,169 (65%)
Noise (unassigned)	4,895 (35%)

The 35% noise rate is typical for HDBSCAN on heterogeneous text corpora — it reflects genuine semantic diversity within the synthetic biology literature rather than a pipeline failure.

City-level patterns

Top 10 cities by total artifact count (papers + projects):

Rank	City	Papers	Projects	Total
1	Beijing	290	258	548
2	Cambridge	294	40	334
3	Shanghai	108	216	324
4	London	236	29	265
5	Tokyo	103	56	159
6	Wuhan	22	134	156
7	Shenzhen	10	145	155
8	Paris	117	37	154
9	New York	98	35	133
10	San Diego	127	0	127

Beijing and Shanghai reflect the rapid growth of Chinese synthetic biology research and iGEM participation since 2012. Cambridge (MA/UK) leads for academic papers, anchored by MIT, Harvard, and the Wellcome Sanger Institute.

Carbon capture subset

267 papers (2.8% of corpus) and 141 iGEM projects (3.1% of corpus) were tagged as carbon-capture related. 30 cities have at least one carbon-capture artifact of both types, providing the basis for city-level cross-type analysis.

Full analysis on the Case Study: Carbon Capture page.

--- title: "Results" --- ## Corpus overview The pipeline currently covers papers and iGEM projects. Patents and parts will be added in a later pipeline run. | Artifact type | Count | Year range | Cities geocoded | |---------------|------:|------------|----------------:| | Papers | 9,449 | 2000–2026 | 8,720 (92%) | | iGEM projects | 4,615 | 2009–2025 | 4,615 (100%) | | Patents | — | — | — | | iGEM parts | — | — | — | | **Total** | **14,064** | **2000–2026** | **13,335 (95%)** | Papers were retrieved from OpenAlex using core synthetic biology keywords ("synthetic biology", "synthetic genomics", "BioBrick") and subfield keywords ("repressilator", "minimal genome", "xenobiology", etc.). iGEM projects were geocoded using a three-tier pipeline: OpenAlex institution lookup → Claude Haiku city extraction → Nominatim coordinates. Paper coordinates were obtained by fetching full institution objects from OpenAlex (which include city and lat/lon) using institution IDs stored during ingestion. ## Semantic space All artifacts were embedded using SPECTER (Lo et al., 2020), a transformer model pre-trained on scientific text to produce 768-dimensional vectors from title and abstract. Embeddings were reduced to 2D with UMAP for visualization. The shared projection shows papers and iGEM projects occupying overlapping but distinguishable regions of semantic space, consistent with the hypothesis that student projects and academic research share the same knowledge domain. See the [Explorer](explorer.qmd) for the interactive version. ## Clustering HDBSCAN was applied to the full 768-dimensional embedding space. | | Count | |---|---:| | Total artifacts | 14,064 | | Clusters found | 256 | | Artifacts assigned to a cluster | 9,169 (65%) | | Noise (unassigned) | 4,895 (35%) | The 35% noise rate is typical for HDBSCAN on heterogeneous text corpora — it reflects genuine semantic diversity within the synthetic biology literature rather than a pipeline failure. ## City-level patterns Top 10 cities by total artifact count (papers + projects): | Rank | City | Papers | Projects | Total | |------|------|-------:|---------:|------:| | 1 | Beijing | 290 | 258 | 548 | | 2 | Cambridge | 294 | 40 | 334 | | 3 | Shanghai | 108 | 216 | 324 | | 4 | London | 236 | 29 | 265 | | 5 | Tokyo | 103 | 56 | 159 | | 6 | Wuhan | 22 | 134 | 156 | | 7 | Shenzhen | 10 | 145 | 155 | | 8 | Paris | 117 | 37 | 154 | | 9 | New York | 98 | 35 | 133 | | 10 | San Diego | 127 | 0 | 127 | Beijing and Shanghai reflect the rapid growth of Chinese synthetic biology research and iGEM participation since 2012. Cambridge (MA/UK) leads for academic papers, anchored by MIT, Harvard, and the Wellcome Sanger Institute. ## Carbon capture subset 267 papers (2.8% of corpus) and 141 iGEM projects (3.1% of corpus) were tagged as carbon-capture related. 30 cities have at least one carbon-capture artifact of both types, providing the basis for city-level cross-type analysis. Full analysis on the [Case Study: Carbon Capture](case-study.qmd) page.